Skip to contentSkip to conversation

mm, swap, frontswap: Fix THP swap if frontswap enabled

PATCH In review Huang, Ying 7 participants 13 messages Feb 7, 2018 – Feb 12, 2018

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mm, swap, frontswap: Fix THP swap if frontswap enabled

From: Huang Ying <[email protected]>

It was reported by Sergey Senozhatsky that if THP (Transparent Huge Page) and frontswap (via zswap) are both enabled, when memory goes low so that swap is triggered, segfault and memory corruption will occur in random user space applications as follow,

kernel: urxvt[338]: segfault at 20 ip 00007fc08889ae0d sp 00007ffc73a7fc40 error 6 in libc-2.26.so[7fc08881a000+1ae000]

 #0  0x00007fc08889ae0d _int_malloc (libc.so.6)
 #1  0x00007fc08889c2f3 malloc (libc.so.6)
 #2  0x0000560e6004bff7 _Z14rxvt_wcstoutf8PKwi (urxvt)
 #3  0x0000560e6005e75c n/a (urxvt)
 #4  0x0000560e6007d9f1 _ZN16rxvt_perl_interp6invokeEP9rxvt_term9hook_typez (urxvt)
 #5  0x0000560e6003d988 _ZN9rxvt_term9cmd_parseEv (urxvt)
 #6  0x0000560e60042804 _ZN9rxvt_term6pty_cbERN2ev2ioEi (urxvt)
 #7  0x0000560e6005c10f _Z17ev_invoke_pendingv (urxvt)
 #8  0x0000560e6005cb55 ev_run (urxvt)
 #9  0x0000560e6003b9b9 main (urxvt)

#10 0x00007fc08883af4a __libc_start_main (libc.so.6) #11 0x0000560e6003f9da _start (urxvt)

After bisection, it was found the first bad commit is bd4c82c22c367e068 ("mm, THP, swap: delay splitting THP after swapped out").

The root cause is as follow.

When the pages are written to swap device during swapping out in swap_writepage(), zswap (fontswap) is tried to compress the pages instead to improve the performance. But zswap (frontswap) will treat THP as normal page, so only the head page is saved. After swapping in, tail pages will not be restored to its original contents, so cause the memory corruption in the applications.

This is fixed via splitting THP before writing the page to swap device if frontswap is enabled. To deal with the situation where frontswap is enabled at runtime, whether the page is THP is checked before using frontswap during swapping out too.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Dan Streetman <[email protected]>
Cc Seth Jennings <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Tetsuo Handa <[email protected]>
Cc Shaohua Li <[email protected]>
Cc Michal Hocko <[email protected]>
Cc Johannes Weiner <[email protected]>
Cc Mel Gorman <[email protected]>
Cc Shakeel Butt <[email protected]>
Fixes bd4c82c22c367e068 ("mm, THP, swap: delay splitting THP after swapped out")

Changelog:

v2:

- Move frontswap check into swapfile.c to avoid to make vmscan.c depends on frontswap.

---
 mm/page_io.c  | 2 +-
 mm/swapfile.c | 3 +++
 2 files changed, 4 insertions(+), 1 deletion(-)
mm/page_io.c +1 -1
index b41cf9644585..6dca817ae7a0 100644
@@ -250,7 +250,7 @@ int swap_writepage(struct page *page, struct writeback_control *wbc)
 		unlock_page(page);
 		goto out;
 	}
-	if (frontswap_store(page) == 0) {
+	if (!PageTransHuge(page) && frontswap_store(page) == 0) {
 		set_page_writeback(page);
 		unlock_page(page);
 		end_page_writeback(page);
mm/swapfile.c +3 -0
index 006047b16814..0b7c7883ce64 100644
@@ -934,6 +934,9 @@ int get_swap_pages(int n_goal, bool cluster, swp_entry_t swp_entries[])
 
 	/* Only single cluster request supported */
 	WARN_ON_ONCE(n_goal > 1 && cluster);
+	/* Frontswap doesn't support THP */
+	if (frontswap_enabled() && cluster)
+		goto noswap;
 
 	avail_pgs = atomic_long_read(&nr_swap_pages) / nr_pages;
 	if (avail_pgs <= 0)
Signature · 7 lines
2.15.1

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Reviewed-by on v1 1/1
#1 lore raw
Quoted 34 lines · Ying

From: Huang Ying <[email protected]>

It was reported by Sergey Senozhatsky that if THP (Transparent Huge Page) and frontswap (via zswap) are both enabled, when memory goes low so that swap is triggered, segfault and memory corruption will occur in random user space applications as follow,

kernel: urxvt[338]: segfault at 20 ip 00007fc08889ae0d sp 00007ffc73a7fc40 error 6 in libc-2.26.so[7fc08881a000+1ae000]

 #0  0x00007fc08889ae0d _int_malloc (libc.so.6)
 #1  0x00007fc08889c2f3 malloc (libc.so.6)
 #2  0x0000560e6004bff7 _Z14rxvt_wcstoutf8PKwi (urxvt)
 #3  0x0000560e6005e75c n/a (urxvt)
 #4  0x0000560e6007d9f1 _ZN16rxvt_perl_interp6invokeEP9rxvt_term9hook_typez (urxvt)
 #5  0x0000560e6003d988 _ZN9rxvt_term9cmd_parseEv (urxvt)
 #6  0x0000560e60042804 _ZN9rxvt_term6pty_cbERN2ev2ioEi (urxvt)
 #7  0x0000560e6005c10f _Z17ev_invoke_pendingv (urxvt)
 #8  0x0000560e6005cb55 ev_run (urxvt)
 #9  0x0000560e6003b9b9 main (urxvt)

#10 0x00007fc08883af4a __libc_start_main (libc.so.6) #11 0x0000560e6003f9da _start (urxvt)

After bisection, it was found the first bad commit is bd4c82c22c367e068 ("mm, THP, swap: delay splitting THP after swapped out").

The root cause is as follow.

When the pages are written to swap device during swapping out in swap_writepage(), zswap (fontswap) is tried to compress the pages instead to improve the performance. But zswap (frontswap) will treat THP as normal page, so only the head page is saved. After swapping in, tail pages will not be restored to its original contents, so cause the memory corruption in the applications.

Quoted 8 lines

This is fixed via splitting THP before writing the page to swap device if frontswap is enabled. To deal with the situation where frontswap is enabled at runtime, whether the page is THP is checked before using frontswap during swapping out too.

Cc Konrad Rzeszutek Wilk <[email protected]>
Quoted 53 lines
Cc Dan Streetman <[email protected]>
Cc Seth Jennings <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Tetsuo Handa <[email protected]>
Cc Shaohua Li <[email protected]>
Cc Michal Hocko <[email protected]>
Cc Johannes Weiner <[email protected]>
Cc Mel Gorman <[email protected]>
Cc Shakeel Butt <[email protected]>
Fixes bd4c82c22c367e068 ("mm, THP, swap: delay splitting THP after swapped out")

Changelog:

v2:

- Move frontswap check into swapfile.c to avoid to make vmscan.c depends on frontswap.

---
 mm/page_io.c  | 2 +-
 mm/swapfile.c | 3 +++
 2 files changed, 4 insertions(+), 1 deletion(-)
mm/page_io.c +1 -1
index b41cf9644585..6dca817ae7a0 100644
@@ -250,7 +250,7 @@ int swap_writepage(struct page *page, struct writeback_control *wbc)
 		unlock_page(page);
 		goto out;
 	}
-	if (frontswap_store(page) == 0) {
+	if (!PageTransHuge(page) && frontswap_store(page) == 0) {
 		set_page_writeback(page);
 		unlock_page(page);
 		end_page_writeback(page);
mm/swapfile.c +3 -0
index 006047b16814..0b7c7883ce64 100644
@@ -934,6 +934,9 @@ int get_swap_pages(int n_goal, bool cluster, swp_entry_t swp_entries[])
 
 	/* Only single cluster request supported */
 	WARN_ON_ONCE(n_goal > 1 && cluster);
+	/* Frontswap doesn't support THP */
+	if (frontswap_enabled() && cluster)
+		goto noswap;
 
 	avail_pgs = atomic_long_read(&nr_swap_pages) / nr_pages;
 	if (avail_pgs <= 0)
Signature · 1 line
2.15.1

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

maintainer
#2 lore raw
In reply to Huang, Ying
Quoted 57 lines · Ying

From: Huang Ying <[email protected]>

It was reported by Sergey Senozhatsky that if THP (Transparent Huge Page) and frontswap (via zswap) are both enabled, when memory goes low so that swap is triggered, segfault and memory corruption will occur in random user space applications as follow,

kernel: urxvt[338]: segfault at 20 ip 00007fc08889ae0d sp 00007ffc73a7fc40 error 6 in libc-2.26.so[7fc08881a000+1ae000]

 #0  0x00007fc08889ae0d _int_malloc (libc.so.6)
 #1  0x00007fc08889c2f3 malloc (libc.so.6)
 #2  0x0000560e6004bff7 _Z14rxvt_wcstoutf8PKwi (urxvt)
 #3  0x0000560e6005e75c n/a (urxvt)
 #4  0x0000560e6007d9f1 _ZN16rxvt_perl_interp6invokeEP9rxvt_term9hook_typez (urxvt)
 #5  0x0000560e6003d988 _ZN9rxvt_term9cmd_parseEv (urxvt)
 #6  0x0000560e60042804 _ZN9rxvt_term6pty_cbERN2ev2ioEi (urxvt)
 #7  0x0000560e6005c10f _Z17ev_invoke_pendingv (urxvt)
 #8  0x0000560e6005cb55 ev_run (urxvt)
 #9  0x0000560e6003b9b9 main (urxvt)

#10 0x00007fc08883af4a __libc_start_main (libc.so.6) #11 0x0000560e6003f9da _start (urxvt)

After bisection, it was found the first bad commit is bd4c82c22c367e068 ("mm, THP, swap: delay splitting THP after swapped out").

The root cause is as follow.

When the pages are written to swap device during swapping out in swap_writepage(), zswap (fontswap) is tried to compress the pages instead to improve the performance. But zswap (frontswap) will treat THP as normal page, so only the head page is saved. After swapping in, tail pages will not be restored to its original contents, so cause the memory corruption in the applications.

This is fixed via splitting THP before writing the page to swap device if frontswap is enabled. To deal with the situation where frontswap is enabled at runtime, whether the page is THP is checked before using frontswap during swapping out too.

...

mm/page_io.c +1 -1
@@ -250,7 +250,7 @@ int swap_writepage(struct page *page, struct writeback_control *wbc)
 		unlock_page(page);
 		goto out;
 	}
-	if (frontswap_store(page) == 0) {
+	if (!PageTransHuge(page) && frontswap_store(page) == 0) {
 		set_page_writeback(page);
 		unlock_page(page);
 		end_page_writeback(page);
mm/swapfile.c +0 -0
index 006047b16814..0b7c7883ce64 100644

mm/swapfile.c int get_swap_pages(int n_goal, bool cluster, swp_entry_t swp_entries[]) · line 940

Quoted 7 lines
mm/swapfile.c +3 -0
@@ -934,6 +934,9 @@ int get_swap_pages(int n_goal, bool cluster, swp_entry_t swp_entries[])
 
 	/* Only single cluster request supported */
 	WARN_ON_ONCE(n_goal > 1 && cluster);
+	/* Frontswap doesn't support THP */
+	if (frontswap_enabled() && cluster)
+		goto noswap;
 

hm. This is assuming that "cluster==true" means "this is thp swap". That's presently true, but is it appropriate that get_swap_pages() is peeking at "cluster" to work out why it is being called?

Or would it be cleaner to do this in get_swap_page()? Something like

mm/swap_slots.c renamed from mm/swap_slots.c~a +5 -2
@@ -317,8 +317,11 @@ swp_entry_t get_swap_page(struct page *p
 	entry.val = 0;
 
 	if (PageTransHuge(page)) {
-		if (IS_ENABLED(CONFIG_THP_SWAP))
-			get_swap_pages(1, true, &entry);
+		/* Frontswap doesn't support THP */
+		if (!frontswap_enabled()) {
+			if (IS_ENABLED(CONFIG_THP_SWAP))
+				get_swap_pages(1, true, &entry);
+		}
 		return entry;
 	}
 

_

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#3 lore raw
Quoted 79 lines · Andrew Morton
Quoted 57 lines · Ying

From: Huang Ying <[email protected]>

It was reported by Sergey Senozhatsky that if THP (Transparent Huge Page) and frontswap (via zswap) are both enabled, when memory goes low so that swap is triggered, segfault and memory corruption will occur in random user space applications as follow,

kernel: urxvt[338]: segfault at 20 ip 00007fc08889ae0d sp 00007ffc73a7fc40 error 6 in libc-2.26.so[7fc08881a000+1ae000]

 #0  0x00007fc08889ae0d _int_malloc (libc.so.6)
 #1  0x00007fc08889c2f3 malloc (libc.so.6)
 #2  0x0000560e6004bff7 _Z14rxvt_wcstoutf8PKwi (urxvt)
 #3  0x0000560e6005e75c n/a (urxvt)
 #4  0x0000560e6007d9f1 _ZN16rxvt_perl_interp6invokeEP9rxvt_term9hook_typez (urxvt)
 #5  0x0000560e6003d988 _ZN9rxvt_term9cmd_parseEv (urxvt)
 #6  0x0000560e60042804 _ZN9rxvt_term6pty_cbERN2ev2ioEi (urxvt)
 #7  0x0000560e6005c10f _Z17ev_invoke_pendingv (urxvt)
 #8  0x0000560e6005cb55 ev_run (urxvt)
 #9  0x0000560e6003b9b9 main (urxvt)

#10 0x00007fc08883af4a __libc_start_main (libc.so.6) #11 0x0000560e6003f9da _start (urxvt)

After bisection, it was found the first bad commit is bd4c82c22c367e068 ("mm, THP, swap: delay splitting THP after swapped out").

The root cause is as follow.

When the pages are written to swap device during swapping out in swap_writepage(), zswap (fontswap) is tried to compress the pages instead to improve the performance. But zswap (frontswap) will treat THP as normal page, so only the head page is saved. After swapping in, tail pages will not be restored to its original contents, so cause the memory corruption in the applications.

This is fixed via splitting THP before writing the page to swap device if frontswap is enabled. To deal with the situation where frontswap is enabled at runtime, whether the page is THP is checked before using frontswap during swapping out too.

...

mm/page_io.c +1 -1
@@ -250,7 +250,7 @@ int swap_writepage(struct page *page, struct writeback_control *wbc)
 		unlock_page(page);
 		goto out;
 	}
-	if (frontswap_store(page) == 0) {
+	if (!PageTransHuge(page) && frontswap_store(page) == 0) {
 		set_page_writeback(page);
 		unlock_page(page);
 		end_page_writeback(page);
mm/swapfile.c +0 -0
index 006047b16814..0b7c7883ce64 100644

mm/swapfile.c int get_swap_pages(int n_goal, bool cluster, swp_entry_t swp_entries[]) · line 940

Quoted 7 lines
mm/swapfile.c +3 -0
@@ -934,6 +934,9 @@ int get_swap_pages(int n_goal, bool cluster, swp_entry_t swp_entries[])
 
 	/* Only single cluster request supported */
 	WARN_ON_ONCE(n_goal > 1 && cluster);
+	/* Frontswap doesn't support THP */
+	if (frontswap_enabled() && cluster)
+		goto noswap;
 

hm. This is assuming that "cluster==true" means "this is thp swap". That's presently true, but is it appropriate that get_swap_pages() is peeking at "cluster" to work out why it is being called?

Or would it be cleaner to do this in get_swap_page()? Something like

mm/swap_slots.c renamed from mm/swap_slots.c~a +0 -0
@@ -317,8 +317,11 @@ swp_entry_t get_swap_page(struct page *p
 	entry.val = 0;
 
 	if (PageTransHuge(page)) {

mm/swap_slots.c swp_entry_t get_swap_page(struct page *p · line 327

Quoted 10 lines
mm/swap_slots.c renamed from mm/swap_slots.c~a +5 -2
-		if (IS_ENABLED(CONFIG_THP_SWAP))
-			get_swap_pages(1, true, &entry);
+		/* Frontswap doesn't support THP */
+		if (!frontswap_enabled()) {
+			if (IS_ENABLED(CONFIG_THP_SWAP))
+				get_swap_pages(1, true, &entry);
+		}
 		return entry;
 	}
 

Sure. I will do this.

Best Regards,
Huang, Ying

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#4 lore raw
In reply to Andrew Morton

On (02/07/18 13:05), Andrew Morton wrote:
[..]

Quoted 11 lines

hm. This is assuming that "cluster==true" means "this is thp swap". That's presently true, but is it appropriate that get_swap_pages() is peeking at "cluster" to work out why it is being called?

Or would it be cleaner to do this in get_swap_page()? Something like

mm/swap_slots.c renamed from mm/swap_slots.c~a +0 -0
@@ -317,8 +317,11 @@ swp_entry_t get_swap_page(struct page *p
 	entry.val = 0;
 

mm/swap_slots.c swp_entry_t get_swap_page(struct page *p · line 326

Quoted 10 lines
mm/swap_slots.c renamed from mm/swap_slots.c~a +5 -2
 	if (PageTransHuge(page)) {
-		if (IS_ENABLED(CONFIG_THP_SWAP))
-			get_swap_pages(1, true, &entry);
+		/* Frontswap doesn't support THP */
+		if (!frontswap_enabled()) {
+			if (IS_ENABLED(CONFIG_THP_SWAP))
+				get_swap_pages(1, true, &entry);
+		}
 		return entry;
 	}

I have proposed exactly the same thing [1], Minchan commented that it would introduce frontswap dependency to swap_slots.c [2]. Which is true, but I'd still probably prefer to handle it all in get_swap_page. Minchan, any objections?

[1] https://marc.info/?l=linux-mm&m=151791052007719&w=2
[2] https://marc.info/?l=linux-mm&m=151792646812617&w=2

	-ss

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#5 lore raw
Quoted 13 lines · Sergey Senozhatsky

On (02/07/18 13:05), Andrew Morton wrote:
[..]

Quoted 11 lines

hm. This is assuming that "cluster==true" means "this is thp swap". That's presently true, but is it appropriate that get_swap_pages() is peeking at "cluster" to work out why it is being called?

Or would it be cleaner to do this in get_swap_page()? Something like

mm/swap_slots.c renamed from mm/swap_slots.c~a +0 -0
@@ -317,8 +317,11 @@ swp_entry_t get_swap_page(struct page *p
 	entry.val = 0;
 
Quoted 15 lines

mm/swap_slots.c swp_entry_t get_swap_page(struct page *p · line 326

Quoted 10 lines
mm/swap_slots.c renamed from mm/swap_slots.c~a +5 -2
 	if (PageTransHuge(page)) {
-		if (IS_ENABLED(CONFIG_THP_SWAP))
-			get_swap_pages(1, true, &entry);
+		/* Frontswap doesn't support THP */
+		if (!frontswap_enabled()) {
+			if (IS_ENABLED(CONFIG_THP_SWAP))
+				get_swap_pages(1, true, &entry);
+		}
 		return entry;
 	}

I have proposed exactly the same thing [1], Minchan commented that it would introduce frontswap dependency to swap_slots.c [2]. Which is true, but I'd still probably prefer to handle it all in get_swap_page. Minchan, any objections?

I didn't want to spread out frontswap stuff unless it has good value because most of frontswap functions are located in mm/swapfile.c at this moment. It gives me good feeling frontswap's abstraction is wonderful.
However, if frontswap matainer has no problem, I am not against, either.

Thanks.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#6 lore raw

On (02/08/18 02:25), Minchan Kim wrote:
[..]

Quoted 16 lines
Quoted 7 lines
Quoted 7 lines
mm/swap_slots.c renamed from mm/swap_slots.c~a +4 -2
 	if (PageTransHuge(page)) {
-		if (IS_ENABLED(CONFIG_THP_SWAP))
-			get_swap_pages(1, true, &entry);
+		/* Frontswap doesn't support THP */
+		if (!frontswap_enabled()) {
+			if (IS_ENABLED(CONFIG_THP_SWAP))
+				get_swap_pages(1, true, &entry);
Quoted 8 lines

mm/swap_slots.c swp_entry_t get_swap_page(struct page *p · line 326

Quoted 3 lines
mm/swap_slots.c renamed from mm/swap_slots.c~a +1 -0
+		}
 		return entry;
 	}

I have proposed exactly the same thing [1], Minchan commented that it would introduce frontswap dependency to swap_slots.c [2]. Which is true, but I'd still probably prefer to handle it all in get_swap_page. Minchan, any objections?

Quoted 3 lines

I didn't want to spread out frontswap stuff unless it has good value because most of frontswap functions are located in mm/swapfile.c at this moment.

Sure, your points are perfectly valid. At the same time it might be the case that we already kind of expose that THP dependency thing to vmscan. The whole

	if (!add_to_swap()) {
		if (!PageTransHuge(page))
			goto activate_locked;

		split_huge_page_to_list(page);
		add_to_swap(page);
	}

looks a bit suspicious - if add_to_swap() fails and the page is THP then split it and add_to_swap() again.

	-ss

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#7 lore raw
In reply to Huang, Ying
Quoted 70 lines · Ying

From: Huang Ying <[email protected]>

It was reported by Sergey Senozhatsky that if THP (Transparent Huge Page) and frontswap (via zswap) are both enabled, when memory goes low so that swap is triggered, segfault and memory corruption will occur in random user space applications as follow,

kernel: urxvt[338]: segfault at 20 ip 00007fc08889ae0d sp 00007ffc73a7fc40 error 6 in libc-2.26.so[7fc08881a000+1ae000]

 #0  0x00007fc08889ae0d _int_malloc (libc.so.6)
 #1  0x00007fc08889c2f3 malloc (libc.so.6)
 #2  0x0000560e6004bff7 _Z14rxvt_wcstoutf8PKwi (urxvt)
 #3  0x0000560e6005e75c n/a (urxvt)
 #4  0x0000560e6007d9f1 _ZN16rxvt_perl_interp6invokeEP9rxvt_term9hook_typez (urxvt)
 #5  0x0000560e6003d988 _ZN9rxvt_term9cmd_parseEv (urxvt)
 #6  0x0000560e60042804 _ZN9rxvt_term6pty_cbERN2ev2ioEi (urxvt)
 #7  0x0000560e6005c10f _Z17ev_invoke_pendingv (urxvt)
 #8  0x0000560e6005cb55 ev_run (urxvt)
 #9  0x0000560e6003b9b9 main (urxvt)

#10 0x00007fc08883af4a __libc_start_main (libc.so.6) #11 0x0000560e6003f9da _start (urxvt)

After bisection, it was found the first bad commit is bd4c82c22c367e068 ("mm, THP, swap: delay splitting THP after swapped out").

The root cause is as follow.

When the pages are written to swap device during swapping out in swap_writepage(), zswap (fontswap) is tried to compress the pages instead to improve the performance. But zswap (frontswap) will treat THP as normal page, so only the head page is saved. After swapping in, tail pages will not be restored to its original contents, so cause the memory corruption in the applications.

This is fixed via splitting THP before writing the page to swap device if frontswap is enabled. To deal with the situation where frontswap is enabled at runtime, whether the page is THP is checked before using frontswap during swapping out too.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Dan Streetman <[email protected]>
Cc Seth Jennings <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Tetsuo Handa <[email protected]>
Cc Shaohua Li <[email protected]>
Cc Michal Hocko <[email protected]>
Cc Johannes Weiner <[email protected]>
Cc Mel Gorman <[email protected]>
Cc Shakeel Butt <[email protected]>
Fixes bd4c82c22c367e068 ("mm, THP, swap: delay splitting THP after swapped out")

Changelog:

v2:

- Move frontswap check into swapfile.c to avoid to make vmscan.c depends on frontswap.

---
 mm/page_io.c  | 2 +-
 mm/swapfile.c | 3 +++
 2 files changed, 4 insertions(+), 1 deletion(-)
mm/page_io.c +0 -0
index b41cf9644585..6dca817ae7a0 100644

mm/page_io.c int swap_writepage(struct page *page, struct writeback_control *wbc) · line 253

Quoted 5 lines
mm/page_io.c +1 -1
@@ -250,7 +250,7 @@ int swap_writepage(struct page *page, struct writeback_control *wbc)
 		unlock_page(page);
 		goto out;
 	}
-	if (frontswap_store(page) == 0) {
+	if (!PageTransHuge(page) && frontswap_store(page) == 0) {

Why do we need this?

If frontswap_enabled is enabled but it doesn't support THP, it doesn't allow cluster allocation by below logic so any THP page shouldn't come this path. What do I missing now?

Quoted 20 lines
mm/page_io.c +0 -0
 		set_page_writeback(page);
 		unlock_page(page);
 		end_page_writeback(page);
mm/swapfile.c +3 -0
index 006047b16814..0b7c7883ce64 100644
@@ -934,6 +934,9 @@ int get_swap_pages(int n_goal, bool cluster, swp_entry_t swp_entries[])
 
 	/* Only single cluster request supported */
 	WARN_ON_ONCE(n_goal > 1 && cluster);
+	/* Frontswap doesn't support THP */
+	if (frontswap_enabled() && cluster)
+		goto noswap;
 
 	avail_pgs = atomic_long_read(&nr_swap_pages) / nr_pages;
 	if (avail_pgs <= 0)
Signature · 1 line
2.15.1

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#8 lore raw
Quoted 79 lines · Minchan Kim
Quoted 70 lines · Ying

From: Huang Ying <[email protected]>

It was reported by Sergey Senozhatsky that if THP (Transparent Huge Page) and frontswap (via zswap) are both enabled, when memory goes low so that swap is triggered, segfault and memory corruption will occur in random user space applications as follow,

kernel: urxvt[338]: segfault at 20 ip 00007fc08889ae0d sp 00007ffc73a7fc40 error 6 in libc-2.26.so[7fc08881a000+1ae000]

 #0  0x00007fc08889ae0d _int_malloc (libc.so.6)
 #1  0x00007fc08889c2f3 malloc (libc.so.6)
 #2  0x0000560e6004bff7 _Z14rxvt_wcstoutf8PKwi (urxvt)
 #3  0x0000560e6005e75c n/a (urxvt)
 #4  0x0000560e6007d9f1 _ZN16rxvt_perl_interp6invokeEP9rxvt_term9hook_typez (urxvt)
 #5  0x0000560e6003d988 _ZN9rxvt_term9cmd_parseEv (urxvt)
 #6  0x0000560e60042804 _ZN9rxvt_term6pty_cbERN2ev2ioEi (urxvt)
 #7  0x0000560e6005c10f _Z17ev_invoke_pendingv (urxvt)
 #8  0x0000560e6005cb55 ev_run (urxvt)
 #9  0x0000560e6003b9b9 main (urxvt)

#10 0x00007fc08883af4a __libc_start_main (libc.so.6) #11 0x0000560e6003f9da _start (urxvt)

After bisection, it was found the first bad commit is bd4c82c22c367e068 ("mm, THP, swap: delay splitting THP after swapped out").

The root cause is as follow.

When the pages are written to swap device during swapping out in swap_writepage(), zswap (fontswap) is tried to compress the pages instead to improve the performance. But zswap (frontswap) will treat THP as normal page, so only the head page is saved. After swapping in, tail pages will not be restored to its original contents, so cause the memory corruption in the applications.

This is fixed via splitting THP before writing the page to swap device if frontswap is enabled. To deal with the situation where frontswap is enabled at runtime, whether the page is THP is checked before using frontswap during swapping out too.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Dan Streetman <[email protected]>
Cc Seth Jennings <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Tetsuo Handa <[email protected]>
Cc Shaohua Li <[email protected]>
Cc Michal Hocko <[email protected]>
Cc Johannes Weiner <[email protected]>
Cc Mel Gorman <[email protected]>
Cc Shakeel Butt <[email protected]>
Fixes bd4c82c22c367e068 ("mm, THP, swap: delay splitting THP after swapped out")

Changelog:

v2:

- Move frontswap check into swapfile.c to avoid to make vmscan.c depends on frontswap.

---
 mm/page_io.c  | 2 +-
 mm/swapfile.c | 3 +++
 2 files changed, 4 insertions(+), 1 deletion(-)
mm/page_io.c +0 -0
index b41cf9644585..6dca817ae7a0 100644

mm/page_io.c int swap_writepage(struct page *page, struct writeback_control *wbc) · line 253

Quoted 5 lines
mm/page_io.c +1 -1
@@ -250,7 +250,7 @@ int swap_writepage(struct page *page, struct writeback_control *wbc)
              unlock_page(page);
              goto out;
      }
-     if (frontswap_store(page) == 0) {
+     if (!PageTransHuge(page) && frontswap_store(page) == 0) {

Why do we need this?

Quoted 3 lines

If frontswap_enabled is enabled but it doesn't support THP, it doesn't allow cluster allocation by below logic so any THP page shouldn't come this path. What do I missing now?

If frontswap_enabled() becomes true at runtime after swap cluster allocation but before swap_writepage(), this can prevent memory corruption. I know this isn't true now. Because frontswap isn't very dynamic now. But I still think this is good thing to do.

Best Regards,
Huang, Ying

Quoted 15 lines
Quoted 15 lines
mm/page_io.c +0 -0
              set_page_writeback(page);
              unlock_page(page);
              end_page_writeback(page);
mm/swapfile.c +3 -0
index 006047b16814..0b7c7883ce64 100644
@@ -934,6 +934,9 @@ int get_swap_pages(int n_goal, bool cluster, swp_entry_t swp_entries[])
 
      /* Only single cluster request supported */
      WARN_ON_ONCE(n_goal > 1 && cluster);
+     /* Frontswap doesn't support THP */
+     if (frontswap_enabled() && cluster)
+             goto noswap;
 
Quoted 5 lines
Quoted 5 lines
mm/swapfile.c +0 -1
      avail_pgs = atomic_long_read(&nr_swap_pages) / nr_pages;
      if (avail_pgs <= 0)
--

2.15.1

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#9 lore raw
In reply to Huang, Ying
Quoted 83 lines · Ying

From: Huang Ying <[email protected]>

It was reported by Sergey Senozhatsky that if THP (Transparent Huge Page) and frontswap (via zswap) are both enabled, when memory goes low so that swap is triggered, segfault and memory corruption will occur in random user space applications as follow,

kernel: urxvt[338]: segfault at 20 ip 00007fc08889ae0d sp 00007ffc73a7fc40 error 6 in libc-2.26.so[7fc08881a000+1ae000]

 #0  0x00007fc08889ae0d _int_malloc (libc.so.6)
 #1  0x00007fc08889c2f3 malloc (libc.so.6)
 #2  0x0000560e6004bff7 _Z14rxvt_wcstoutf8PKwi (urxvt)
 #3  0x0000560e6005e75c n/a (urxvt)
 #4  0x0000560e6007d9f1 _ZN16rxvt_perl_interp6invokeEP9rxvt_term9hook_typez (urxvt)
 #5  0x0000560e6003d988 _ZN9rxvt_term9cmd_parseEv (urxvt)
 #6  0x0000560e60042804 _ZN9rxvt_term6pty_cbERN2ev2ioEi (urxvt)
 #7  0x0000560e6005c10f _Z17ev_invoke_pendingv (urxvt)
 #8  0x0000560e6005cb55 ev_run (urxvt)
 #9  0x0000560e6003b9b9 main (urxvt)

#10 0x00007fc08883af4a __libc_start_main (libc.so.6) #11 0x0000560e6003f9da _start (urxvt)

After bisection, it was found the first bad commit is bd4c82c22c367e068 ("mm, THP, swap: delay splitting THP after swapped out").

The root cause is as follow.

When the pages are written to swap device during swapping out in swap_writepage(), zswap (fontswap) is tried to compress the pages instead to improve the performance. But zswap (frontswap) will treat THP as normal page, so only the head page is saved. After swapping in, tail pages will not be restored to its original contents, so cause the memory corruption in the applications.

This is fixed via splitting THP before writing the page to swap device if frontswap is enabled. To deal with the situation where frontswap is enabled at runtime, whether the page is THP is checked before using frontswap during swapping out too.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Dan Streetman <[email protected]>
Cc Seth Jennings <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Tetsuo Handa <[email protected]>
Cc Shaohua Li <[email protected]>
Cc Michal Hocko <[email protected]>
Cc Johannes Weiner <[email protected]>
Cc Mel Gorman <[email protected]>
Cc Shakeel Butt <[email protected]>
Fixes bd4c82c22c367e068 ("mm, THP, swap: delay splitting THP after swapped out")

Changelog:

v2:

- Move frontswap check into swapfile.c to avoid to make vmscan.c depends on frontswap.

---
 mm/page_io.c  | 2 +-
 mm/swapfile.c | 3 +++
 2 files changed, 4 insertions(+), 1 deletion(-)
mm/page_io.c +1 -1
index b41cf9644585..6dca817ae7a0 100644
@@ -250,7 +250,7 @@ int swap_writepage(struct page *page, struct writeback_control *wbc)
                unlock_page(page);
                goto out;
        }
-       if (frontswap_store(page) == 0) {
+       if (!PageTransHuge(page) && frontswap_store(page) == 0) {
                set_page_writeback(page);
                unlock_page(page);
                end_page_writeback(page);
mm/swapfile.c +0 -0
index 006047b16814..0b7c7883ce64 100644

mm/swapfile.c int get_swap_pages(int n_goal, bool cluster, swp_entry_t swp_entries[]) · line 939

Quoted 6 lines
mm/swapfile.c +3 -0
@@ -934,6 +934,9 @@ int get_swap_pages(int n_goal, bool cluster, swp_entry_t swp_entries[])
 
        /* Only single cluster request supported */
        WARN_ON_ONCE(n_goal > 1 && cluster);
+       /* Frontswap doesn't support THP */
+       if (frontswap_enabled() && cluster)
+               goto noswap;

I found this will cause THP swap optimization be turned off forever if CONFIG_ZSWAP=y (which cannot =m). Because frontswap is enabled quite statically instead of dynamically. If frontswap_ops is registered, it will be enabled unconditionally and forever. And zswap will register frontswap_ops during initialize regardless whether zswap is enabled or not.

So I think it will be better to remove swapfile.c changes in this patch, just keep page_io.c changes. Because THP is more dynamic, it is usually used if madvised by default. And even if it is used always by default, this can be changed dynamically. And even THP is used, zswap can still be used for non-THP pages.

Best Regards,
Huang, Ying

Quoted 6 lines
mm/swapfile.c +0 -1
        avail_pgs = atomic_long_read(&nr_swap_pages) / nr_pages;
        if (avail_pgs <= 0)
--

2.15.1

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#10 lore raw

Hi Huang,

Quoted 85 lines · huang ying
Quoted 84 lines · Ying

From: Huang Ying <[email protected]>

It was reported by Sergey Senozhatsky that if THP (Transparent Huge Page) and frontswap (via zswap) are both enabled, when memory goes low so that swap is triggered, segfault and memory corruption will occur in random user space applications as follow,

kernel: urxvt[338]: segfault at 20 ip 00007fc08889ae0d sp 00007ffc73a7fc40 error 6 in libc-2.26.so[7fc08881a000+1ae000]

 #0  0x00007fc08889ae0d _int_malloc (libc.so.6)
 #1  0x00007fc08889c2f3 malloc (libc.so.6)
 #2  0x0000560e6004bff7 _Z14rxvt_wcstoutf8PKwi (urxvt)
 #3  0x0000560e6005e75c n/a (urxvt)
 #4  0x0000560e6007d9f1 _ZN16rxvt_perl_interp6invokeEP9rxvt_term9hook_typez (urxvt)
 #5  0x0000560e6003d988 _ZN9rxvt_term9cmd_parseEv (urxvt)
 #6  0x0000560e60042804 _ZN9rxvt_term6pty_cbERN2ev2ioEi (urxvt)
 #7  0x0000560e6005c10f _Z17ev_invoke_pendingv (urxvt)
 #8  0x0000560e6005cb55 ev_run (urxvt)
 #9  0x0000560e6003b9b9 main (urxvt)

#10 0x00007fc08883af4a __libc_start_main (libc.so.6) #11 0x0000560e6003f9da _start (urxvt)

After bisection, it was found the first bad commit is bd4c82c22c367e068 ("mm, THP, swap: delay splitting THP after swapped out").

The root cause is as follow.

When the pages are written to swap device during swapping out in swap_writepage(), zswap (fontswap) is tried to compress the pages instead to improve the performance. But zswap (frontswap) will treat THP as normal page, so only the head page is saved. After swapping in, tail pages will not be restored to its original contents, so cause the memory corruption in the applications.

This is fixed via splitting THP before writing the page to swap device if frontswap is enabled. To deal with the situation where frontswap is enabled at runtime, whether the page is THP is checked before using frontswap during swapping out too.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Dan Streetman <[email protected]>
Cc Seth Jennings <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Tetsuo Handa <[email protected]>
Cc Shaohua Li <[email protected]>
Cc Michal Hocko <[email protected]>
Cc Johannes Weiner <[email protected]>
Cc Mel Gorman <[email protected]>
Cc Shakeel Butt <[email protected]>
Fixes bd4c82c22c367e068 ("mm, THP, swap: delay splitting THP after swapped out")

Changelog:

v2:

- Move frontswap check into swapfile.c to avoid to make vmscan.c depends on frontswap.

---
 mm/page_io.c  | 2 +-
 mm/swapfile.c | 3 +++
 2 files changed, 4 insertions(+), 1 deletion(-)
mm/page_io.c +1 -1
index b41cf9644585..6dca817ae7a0 100644
@@ -250,7 +250,7 @@ int swap_writepage(struct page *page, struct writeback_control *wbc)
                unlock_page(page);
                goto out;
        }
-       if (frontswap_store(page) == 0) {
+       if (!PageTransHuge(page) && frontswap_store(page) == 0) {
                set_page_writeback(page);
                unlock_page(page);
                end_page_writeback(page);
mm/swapfile.c +0 -0
index 006047b16814..0b7c7883ce64 100644
@@ -934,6 +934,9 @@ int get_swap_pages(int n_goal, bool cluster, swp_entry_t swp_entries[])
 
Quoted 12 lines

mm/swapfile.c int get_swap_pages(int n_goal, bool cluster, swp_entry_t swp_entries[]) · line 939

Quoted 5 lines
mm/swapfile.c +3 -0
        /* Only single cluster request supported */
        WARN_ON_ONCE(n_goal > 1 && cluster);
+       /* Frontswap doesn't support THP */
+       if (frontswap_enabled() && cluster)
+               goto noswap;

I found this will cause THP swap optimization be turned off forever if CONFIG_ZSWAP=y (which cannot =m). Because frontswap is enabled quite statically instead of dynamically. If frontswap_ops is registered, it will be enabled unconditionally and forever. And zswap will register frontswap_ops during initialize regardless whether zswap is enabled or not.

Indeed.

Quoted 3 lines

So I think it will be better to remove swapfile.c changes in this patch, just keep page_io.c changes. Because THP is more dynamic, it

Then, I think it should be done by frontswap backend rather than generic swap layer. Because there are two backends now and one of them can support first.

drivers/xen/tmem.c +3 -0
index bf13d1ec51f3..bdaf309aeea6 100644
@@ -284,6 +284,9 @@ static int tmem_frontswap_store(unsigned type, pgoff_t offset,
        int pool = tmem_frontswap_poolid;
        int ret;
 
+       if (PageTransHuge(page))
+               return -EINVAL;
+
        if (pool < 0)
                return -1;
        if (ind64 != ind)
mm/zswap.c +3 -0
index c004aa4fd3f4..e343534d2892 100644
@@ -1007,6 +1007,9 @@ static int zswap_frontswap_store(unsigned type, pgoff_t offset,
        u8 *src, *dst;
        struct zswap_header zhdr = { .swpentry = swp_entry(type, offset) };
 
+       if (PageTransHuge(page))
+               return -EINVAL;
+
        if (!zswap_enabled || !tree) {
                ret = -ENODEV;
                goto reject;

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#11 lore raw
Quoted 130 lines · Minchan Kim

Hi Huang,

Quoted 85 lines · huang ying
Quoted 84 lines · Ying

From: Huang Ying <[email protected]>

It was reported by Sergey Senozhatsky that if THP (Transparent Huge Page) and frontswap (via zswap) are both enabled, when memory goes low so that swap is triggered, segfault and memory corruption will occur in random user space applications as follow,

kernel: urxvt[338]: segfault at 20 ip 00007fc08889ae0d sp 00007ffc73a7fc40 error 6 in libc-2.26.so[7fc08881a000+1ae000]

 #0  0x00007fc08889ae0d _int_malloc (libc.so.6)
 #1  0x00007fc08889c2f3 malloc (libc.so.6)
 #2  0x0000560e6004bff7 _Z14rxvt_wcstoutf8PKwi (urxvt)
 #3  0x0000560e6005e75c n/a (urxvt)
 #4  0x0000560e6007d9f1 _ZN16rxvt_perl_interp6invokeEP9rxvt_term9hook_typez (urxvt)
 #5  0x0000560e6003d988 _ZN9rxvt_term9cmd_parseEv (urxvt)
 #6  0x0000560e60042804 _ZN9rxvt_term6pty_cbERN2ev2ioEi (urxvt)
 #7  0x0000560e6005c10f _Z17ev_invoke_pendingv (urxvt)
 #8  0x0000560e6005cb55 ev_run (urxvt)
 #9  0x0000560e6003b9b9 main (urxvt)

#10 0x00007fc08883af4a __libc_start_main (libc.so.6) #11 0x0000560e6003f9da _start (urxvt)

After bisection, it was found the first bad commit is bd4c82c22c367e068 ("mm, THP, swap: delay splitting THP after swapped out").

The root cause is as follow.

When the pages are written to swap device during swapping out in swap_writepage(), zswap (fontswap) is tried to compress the pages instead to improve the performance. But zswap (frontswap) will treat THP as normal page, so only the head page is saved. After swapping in, tail pages will not be restored to its original contents, so cause the memory corruption in the applications.

This is fixed via splitting THP before writing the page to swap device if frontswap is enabled. To deal with the situation where frontswap is enabled at runtime, whether the page is THP is checked before using frontswap during swapping out too.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Dan Streetman <[email protected]>
Cc Seth Jennings <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Tetsuo Handa <[email protected]>
Cc Shaohua Li <[email protected]>
Cc Michal Hocko <[email protected]>
Cc Johannes Weiner <[email protected]>
Cc Mel Gorman <[email protected]>
Cc Shakeel Butt <[email protected]>
Fixes bd4c82c22c367e068 ("mm, THP, swap: delay splitting THP after swapped out")

Changelog:

v2:

- Move frontswap check into swapfile.c to avoid to make vmscan.c depends on frontswap.

---
 mm/page_io.c  | 2 +-
 mm/swapfile.c | 3 +++
 2 files changed, 4 insertions(+), 1 deletion(-)
mm/page_io.c +1 -1
index b41cf9644585..6dca817ae7a0 100644
@@ -250,7 +250,7 @@ int swap_writepage(struct page *page, struct writeback_control *wbc)
                unlock_page(page);
                goto out;
        }
-       if (frontswap_store(page) == 0) {
+       if (!PageTransHuge(page) && frontswap_store(page) == 0) {
                set_page_writeback(page);
                unlock_page(page);
                end_page_writeback(page);
mm/swapfile.c +0 -0
index 006047b16814..0b7c7883ce64 100644
@@ -934,6 +934,9 @@ int get_swap_pages(int n_goal, bool cluster, swp_entry_t swp_entries[])
 
Quoted 12 lines

mm/swapfile.c int get_swap_pages(int n_goal, bool cluster, swp_entry_t swp_entries[]) · line 939

Quoted 5 lines
mm/swapfile.c +3 -0
        /* Only single cluster request supported */
        WARN_ON_ONCE(n_goal > 1 && cluster);
+       /* Frontswap doesn't support THP */
+       if (frontswap_enabled() && cluster)
+               goto noswap;

I found this will cause THP swap optimization be turned off forever if CONFIG_ZSWAP=y (which cannot =m). Because frontswap is enabled quite statically instead of dynamically. If frontswap_ops is registered, it will be enabled unconditionally and forever. And zswap will register frontswap_ops during initialize regardless whether zswap is enabled or not.

Indeed.

Quoted 3 lines

So I think it will be better to remove swapfile.c changes in this patch, just keep page_io.c changes. Because THP is more dynamic, it

Then, I think it should be done by frontswap backend rather than generic swap layer. Because there are two backends now and one of them can support first.

drivers/xen/tmem.c +3 -0
index bf13d1ec51f3..bdaf309aeea6 100644
@@ -284,6 +284,9 @@ static int tmem_frontswap_store(unsigned type, pgoff_t offset,
        int pool = tmem_frontswap_poolid;
        int ret;
 
+       if (PageTransHuge(page))
+               return -EINVAL;
+
        if (pool < 0)
                return -1;
        if (ind64 != ind)
mm/zswap.c +0 -0
index c004aa4fd3f4..e343534d2892 100644

mm/zswap.c static int zswap_frontswap_store(unsigned type, pgoff_t offset, · line 1015

Quoted 9 lines
mm/zswap.c +3 -0
@@ -1007,6 +1007,9 @@ static int zswap_frontswap_store(unsigned type, pgoff_t offset,
        u8 *src, *dst;
        struct zswap_header zhdr = { .swpentry = swp_entry(type, offset) };
 
+       if (PageTransHuge(page))
+               return -EINVAL;
+
        if (!zswap_enabled || !tree) {
                ret = -ENODEV;
                goto reject;

Good suggestion! I will do this.

Best Regards,
Huang, Ying

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#12 lore raw
In reply to Minchan Kim
Quoted 104 lines · Minchan Kim

Hi Huang,

Quoted 85 lines · huang ying
Quoted 84 lines · Ying

From: Huang Ying <[email protected]>

It was reported by Sergey Senozhatsky that if THP (Transparent Huge Page) and frontswap (via zswap) are both enabled, when memory goes low so that swap is triggered, segfault and memory corruption will occur in random user space applications as follow,

kernel: urxvt[338]: segfault at 20 ip 00007fc08889ae0d sp 00007ffc73a7fc40 error 6 in libc-2.26.so[7fc08881a000+1ae000]

 #0  0x00007fc08889ae0d _int_malloc (libc.so.6)
 #1  0x00007fc08889c2f3 malloc (libc.so.6)
 #2  0x0000560e6004bff7 _Z14rxvt_wcstoutf8PKwi (urxvt)
 #3  0x0000560e6005e75c n/a (urxvt)
 #4  0x0000560e6007d9f1 _ZN16rxvt_perl_interp6invokeEP9rxvt_term9hook_typez (urxvt)
 #5  0x0000560e6003d988 _ZN9rxvt_term9cmd_parseEv (urxvt)
 #6  0x0000560e60042804 _ZN9rxvt_term6pty_cbERN2ev2ioEi (urxvt)
 #7  0x0000560e6005c10f _Z17ev_invoke_pendingv (urxvt)
 #8  0x0000560e6005cb55 ev_run (urxvt)
 #9  0x0000560e6003b9b9 main (urxvt)

#10 0x00007fc08883af4a __libc_start_main (libc.so.6) #11 0x0000560e6003f9da _start (urxvt)

After bisection, it was found the first bad commit is bd4c82c22c367e068 ("mm, THP, swap: delay splitting THP after swapped out").

The root cause is as follow.

When the pages are written to swap device during swapping out in swap_writepage(), zswap (fontswap) is tried to compress the pages instead to improve the performance. But zswap (frontswap) will treat THP as normal page, so only the head page is saved. After swapping in, tail pages will not be restored to its original contents, so cause the memory corruption in the applications.

This is fixed via splitting THP before writing the page to swap device if frontswap is enabled. To deal with the situation where frontswap is enabled at runtime, whether the page is THP is checked before using frontswap during swapping out too.

Cc Konrad Rzeszutek Wilk <[email protected]>
Cc Dan Streetman <[email protected]>
Cc Seth Jennings <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Tetsuo Handa <[email protected]>
Cc Shaohua Li <[email protected]>
Cc Michal Hocko <[email protected]>
Cc Johannes Weiner <[email protected]>
Cc Mel Gorman <[email protected]>
Cc Shakeel Butt <[email protected]>
Fixes bd4c82c22c367e068 ("mm, THP, swap: delay splitting THP after swapped out")

Changelog:

v2:

- Move frontswap check into swapfile.c to avoid to make vmscan.c depends on frontswap.

---
 mm/page_io.c  | 2 +-
 mm/swapfile.c | 3 +++
 2 files changed, 4 insertions(+), 1 deletion(-)
mm/page_io.c +1 -1
index b41cf9644585..6dca817ae7a0 100644
@@ -250,7 +250,7 @@ int swap_writepage(struct page *page, struct writeback_control *wbc)
                unlock_page(page);
                goto out;
        }
-       if (frontswap_store(page) == 0) {
+       if (!PageTransHuge(page) && frontswap_store(page) == 0) {
                set_page_writeback(page);
                unlock_page(page);
                end_page_writeback(page);
mm/swapfile.c +0 -0
index 006047b16814..0b7c7883ce64 100644
@@ -934,6 +934,9 @@ int get_swap_pages(int n_goal, bool cluster, swp_entry_t swp_entries[])
 
Quoted 12 lines

mm/swapfile.c int get_swap_pages(int n_goal, bool cluster, swp_entry_t swp_entries[]) · line 939

Quoted 5 lines
mm/swapfile.c +3 -0
        /* Only single cluster request supported */
        WARN_ON_ONCE(n_goal > 1 && cluster);
+       /* Frontswap doesn't support THP */
+       if (frontswap_enabled() && cluster)
+               goto noswap;

I found this will cause THP swap optimization be turned off forever if CONFIG_ZSWAP=y (which cannot =m). Because frontswap is enabled quite statically instead of dynamically. If frontswap_ops is registered, it will be enabled unconditionally and forever. And zswap will register frontswap_ops during initialize regardless whether zswap is enabled or not.

Indeed.

Quoted 1 line
Quoted 6 lines
Quoted 2 lines

So I think it will be better to remove swapfile.c changes in this patch, just keep page_io.c changes. Because THP is more dynamic, it

Then, I think it should be done by frontswap backend rather than generic swap layer. Because there are two backends now and one of them can support first.

I like this approach. It allows zswap or xen tmem to support THP in the future.

Quoted 22 lines
drivers/xen/tmem.c +3 -0
index bf13d1ec51f3..bdaf309aeea6 100644
@@ -284,6 +284,9 @@ static int tmem_frontswap_store(unsigned type, pgoff_t offset,
        int pool = tmem_frontswap_poolid;
        int ret;
 
+       if (PageTransHuge(page))
+               return -EINVAL;
+
        if (pool < 0)
                return -1;
        if (ind64 != ind)
mm/zswap.c +0 -0
index c004aa4fd3f4..e343534d2892 100644

mm/zswap.c static int zswap_frontswap_store(unsigned type, pgoff_t offset, · line 1015

Quoted 9 lines
mm/zswap.c +3 -0
@@ -1007,6 +1007,9 @@ static int zswap_frontswap_store(unsigned type, pgoff_t offset,
        u8 *src, *dst;
        struct zswap_header zhdr = { .swpentry = swp_entry(type, offset) };
 
+       if (PageTransHuge(page))
+               return -EINVAL;
+
        if (!zswap_enabled || !tree) {
                ret = -ENODEV;
                goto reject;

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>